The Merchant's Charter — for the small startup
Run ONE great open model as cheaply as it can honestly be run.
How we arrived at this counsel
DeepSeek V4 Flash is the smallest of the three flagship models we advise on, so it's the cheapest model in this class to serve. Its 284B parameters need at least 340.8 GB of GPU memory (284 x 1GB x 1.2). No single card gets close, and desktop/workstation cards can't pool memory across units — so the cheapest honest option is the smallest real NVLink cluster that clears the number: the 4-GPU H200 HGX server at 564 GB pooled.
Model: DeepSeek V4 Flash (284B parameters)
Minimum GPU memory = 284 × 1 GB × 1.2 = 340.8 GB
Our recommended engine
Inquire & Purchase →
H200 HGX 4-GPU NVLink Server
| Memory | 564 GB (NVLink-pooled) |
|---|---|
| Power drawn | 2800 W — roughly 2.33 homes' worth, continuously |
| Total price | $140,000 |